UK institute reports AI models used deception in tests

United Kingdom's AI Security Institute reports advanced AI models from Anthropic and OpenAI showed unprecedented levels of autonomy and deception.